Hi Shant,
3dSeg has not been well documented because it is still in a fluid state. I should be adding the help for it sometime soon. In the meanwhile, here are some answers to your queries.
1- You are correct about the prefix usage. All output volumes end up in the directory specified by 'prefix'.
2- The classifications are in the dataset classes+orig. To access the classes, the safest way is by using the labels directly:
3dcalc -a Classes+orig'<WM>' -expr 'step(a)' -prefix WM_only
or you can do:
3dcalc -a Classes+orig -expr 'equals(a,3)' -prefix WM_only
3- As for the other volumes: Anat is the original input, AnatUB is the input corrected for bias field, Posterior is the posterior probability of class belonging given the estimated distributions. The Unmodulated.p and .c datasets are like Posterior and Classes when prior probabilities are set to a constant after distribution parameters are obtained.
4- Note that in the absence of spatial- or texture-derived priors 3dSeg needs brain mask (or a masked input). How spatial or texture-derived priors can be added will be detailed in the near future, I hope.
cheers,
Ziad